In view of the low efficiency found in the traditional fault prediction methods to predict rolling bearing faults under different working conditions, a new fault prediction method of rolling bearing based on BP neural network and DS evidence theory has thus been proposed. Firstly the wavelet packet decomposition, which is good at processing non-stationary signals, is used to analyze the characteristics of the original vibration data collected by multiple sensors. Next, the structure and parameters of BP neural network are optimized, with multiple BP neural networks used to train the fault prediction model respectively. And finally, the DS evidence theory is used to fuse the prediction results obtained by the multiple neural networks with the final prediction result output. The experimental results show that the proposed method can effectively predict the fault of rolling bearing under different working conditions, with the average accuracy of fault prediction attaining 96.37%. Compared with the methods proposed in the related literature, the accuracy of the rolling bearing fault prediction obtained by the proposed method has been improved.